Calcium deblooming in coronary computed tomography angiography via semantic-oriented generative adversarial network

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Cardiovascular Diagnostics

Background:

• Calcium blooming artifacts in coronary computed tomography angiography (CCTA) lead to false positives in stenosis diagnosis.
• Existing noise reduction techniques are insufficient for mitigating these specific artifacts.
• Calcified plaques are a primary source of blooming artifacts, necessitating targeted solutions.

Purpose of the Study:

• To develop an automated, robust, semantics-oriented adversarial network for CCTA artifact reduction.
• To specifically address and minimize calcium blooming artifacts originating from plaque calcification.
• To improve the diagnostic accuracy of CCTA by enhancing image quality and reducing false positives.

Main Methods:

• Designed a novel semantics-oriented adversarial network leveraging calcified plaques as semantic regions.
• Employed a feature extraction module, global-local fusion, and a generator with a semantic similarity module.
• Validated the network on virtual and a clinical dataset (372 CCTA/CAG pairs) with expert radiologist evaluation.

Main Results:

• The proposed network effectively reduced calcium blooming artifacts in the three major coronary arteries.
• Significant improvements were observed in specificity and positive predictive value for coronary stenosis diagnosis.
• Clinical evaluation confirmed the method's effectiveness in real-world CCTA data.

Conclusions:

• The developed semantics-oriented adversarial network offers a powerful solution for CCTA calcium blooming artifacts.
• This approach enhances diagnostic reliability and accuracy in identifying coronary artery stenosis.
• The method shows promise for routine clinical application in cardiovascular imaging.